Metrics radtorch.metrics

ClassifierMetrics

ClassifierMetrics class a set of methods that enables quantiative evaluation of a trained image classifier performance.

Parameters:

Name	Type	Description	Default
classifier	ImageClassifier	trained ImageClassifier.	required
use_best_model	bool	True to use the model with the lowest validation loss.	required
device	str	Device to be used for training. Default: 'auto' which automtically detects GPU presence and uses it for feature extraction. Options: 'auto', 'cuda', 'cpu'.	'auto'

Methods

classification_report(self, subset='test')

Returns text report showing the main classification metrics.

Parameters:

Name	Type	Description	Default
subset	str	subset of the dataset to be used. This can be 'train', 'valid', or 'test'.	'test'

confusion_matrix(self, subset='test', figsize=(8, 6), cmap='Blues', percent=False)

Returns confusion matrix using target data. Code Adapted from https://github.com/DTrimarchi10/confusion_matrix/blob/master/cf_matrix.py

Parameters:

Name	Type	Description	Default
subset	str	subset of the dataset to be used. This can be 'train', 'valid', or 'test'.	'test'
figsize	tuple	size of the displayed figure.	(8, 6)
cmap	string	Name of Matplotlib color map to be used. See Matplotlib colormaps	'Blues'
percent	bool	True to use percentages instead of real values.	False

Returns:

Type	Description
figure	figure containing confusion matrix

get_predictions(self, subset)

new dataframe is created : self.pred_table, with important columns : label_id and pred_id

roc(self, subset='test', figure_size=(8, 6), plot=True)

Displays ROC of the trained classifier and returns ROC-AUC.

Parameters:

Name	Type	Description	Default
subset	str	subset of the dataset to be used. This can be 'train', 'valid', or 'test'.	'test'
figsize	tuple	size of the displayed figure.	required
plot	bool	True to display ROC.	True

Returns:

Type	Description
float	float of ROC-AUC